基于电子鼻技术的膀胱癌尿液VOCs标志物气体检测与识别  

Detection and recognition of urinary VOCs marker gases for bladder cancer based on electronic nose technology

作　　者：黄志健 韩雨彤 孙宇帆 朱志刚 Huang Zhijian;Han Yutong;Sun Yufan;Zhu Zhigang(School of Health Science and Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学健康科学与工程学院,上海200093

出　　处：《国际生物医学工程杂志》2024年第2期115-122,共8页International Journal of Biomedical Engineering

基　　金：上海市科委高校能力建设(21010502800)。

摘　　要：目的设计一款电子鼻,用于检测与识别膀胱癌尿液中的挥发性有机化合物(VOCs)标志物气体。方法选取异丙醇、乙苯、乙酸和氨气作为目标气体,由8款金属氧化物气体传感器构建传感器阵列进行测试收集实验数据,并对不同特征归一化处理。通过递归特征消除(RFE)筛选出最佳特征子集,进一步引入主成分分析(PCA)和线性判别分析(LDA)降低数据维度便于可视化分析。此外,结合支持向量机(SVM)、随机森林(RF)、K最近邻(KNN)3种机器学习算法进行模型训练和验证。结果特征数为12时,模型分类的准确率最高,特征子集由5个差值、5个灵敏度和2个积分组成,同时将数据降至12维;仅PCA无法区分4种气体,LDA分类效果明显好于PCA,除异丙醇与乙酸有小部分重叠区域,能够将乙苯、氨气很好地与前二者区分开,且样本点聚集在一起,聚类效果也更佳。SVM、RF和KNN的预测准确率分别为0.85、0.56、0.79,经过模型验证,PCA+SVM、LDA+RF和LDA+KNN的分类准确率分别为0.97、0.94、0.97。结论设计了一款电子鼻,能够用于检测与识别膀胱癌尿液VOCs标志物气体。Objective To design an electronic nose that can detect and identify urinary volatile organic compounds(VOCs)as marker gases for bladder cancer.Methods Isopropyl alcohol,ethylbenzene,acetic acid,and ammonia were selected as target gases,and 8 metal oxide gas sensors were used to construct sensor arrays for testing and collecting experimental data,and different characteristics were normalized.Recursive feature elimination(RFE)was used to select the best feature subset,and principal component analysis(PCA)and linear discriminant analysis(LDA)were further introduced to reduce the data dimension and facilitate visual analysis.In addition,three machine learning algorithms,including support vector machine(SVM),random forest(RF),and K-nearest neighbor(KNN),were combined to train and verify the model.Results When the feature number was 12,the accuracy of the model classification had the best performance.The feature subset consisted of 5 differences,5 sensitivities,and 2 integrals,and the data was reduced to 12 dimensions.Only PCA couldn’t distinguish the four gases.The LDA classification performance was significantly better than that of PCA,except that isopropyl alcohol and acetic acid had a small overlap area.LDA could distinguish ethylbenzene and ammonia from isopropyl alcohol and acetic acid;the sample points were gathered,which means the clustering performance was also better.The prediction accuracy of SVM,RF,and KNN was 0.85,0.56,and 0.79,respectively.After model verification,the classification accuracy of PCA+SVM,LDA+RF,and LDA+KNN was 0.97,0.94,and 0.97,respectively.Conclusions An electronic nose was designed to detect and identify urinary VOCs marker gases for bladder cancer.

关 键 词：电子鼻 膀胱癌 挥发性有机化合物 特征工程 气体识别 

分 类 号：R737.14[医药卫生—肿瘤]